I've Changed My Mind: Robots Adapting to Changing Human Goals during Collaboration

2025-11-19

AI Summary

To address the challenge of dynamically changing human goals in human-robot collaboration—where existing methods struggle to adapt in real time without explicit communication—this paper proposes an active inference-based receding-horizon adaptive planning framework. The method integrates a multi-hypothesis behavioral tracking mechanism with a policy library for plausibility validation, a goal-change detection module, and online belief-state updating. Crucially, it employs a discriminative action selection strategy that actively executes, within a receding-horizon planning tree, actions most informative for inferring the human’s updated goal. Evaluated in a cooking simulation environment comprising 30 recipes, the approach achieves significant improvements over three baseline methods: +21.3% in goal recognition accuracy and a 34.7% reduction in average task completion time. These results demonstrate its effectiveness in dynamic goal adaptation and flexible human-robot coordination.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Intelligent Robots: Human-Robot Interaction

Application Category

• Economics, Online Markets and Human Computation: Research challenges in human and human-AI computation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

For effective human-robot collaboration, a robot must align its actions with human goals, even as they change mid-task. Prior approaches often assume fixed goals, reducing goal prediction to a one-time inference. However, in real-world scenarios, humans frequently shift goals, making it challenging for robots to adapt without explicit communication. We propose a method for detecting goal changes by tracking multiple candidate action sequences and verifying their plausibility against a policy bank. Upon detecting a change, the robot refines its belief in relevant past actions and constructs Receding Horizon Planning (RHP) trees to actively select actions that assist the human while encouraging Differentiating Actions to reveal their updated goal. We evaluate our approach in a collaborative cooking environment with up to 30 unique recipes and compare it to three comparable human goal prediction algorithms. Our method outperforms all baselines, quickly converging to the correct goal after a switch, reducing task completion time, and improving collaboration efficiency.

Problem

• Robots must adapt to dynamically changing human goals during collaboration
• Existing methods fail when humans shift goals mid-task without communication
• Proposed system detects goal changes and replans actions to assist humans

Innovation

• Tracking multiple candidate action sequences for goal detection
• Using Receding Horizon Planning trees for action selection
• Employing Differentiating Actions to reveal updated human goals